Method for bilateral filtering of digital images

Abstract

A method for processing an image using a bilateral filter. The bilateral filter is reformulated at each pixel location in the image into a sum of the original signal value of a central pixel at said pixel location and a bilateral correction term which is a function of local signal differences between the central pixel and its neighbors. The bilateral correction term is calculated using a computationally efficient approximation.

Description

BACKGROUND The bilateral filter is a non-linear filter introduced by Tomasi and Manduchi (see “Bilateral filtering for gray and color images”, Proc. IEE Intl. Conf. on Computer Vision, Bombay, India, 1998), which is used for selective denoising of images without blurring image edges. The bilateral filter is based on the idea of modifying the weights of a convolution mask in an image dependent manner, as a function of intensity differences between a pixel under consideration and its neighbors. While the original convolution weights sum to one (as required from any mean-preserving filter), the sum of the modified weights is less than one, so that in order to retain the local normalization of the modified convolution weights, the convolution result is divided by the sum of the modified weights. This single division operation considerably increases the computational complexity necessary to implement a bilateral filter with a digital processor. In addition, in hardware implementations i...

AI Technical Summary

Problems solved by technology

This single division operation considerably increases the computational complexity necessary to implement a bilateral filter with a digital processor.

In addition, in hardware implementations it forces the inclusion of an integer division circuit which would be otherwise unnecessary.

In implementing a bilateral filter, the DIVIDE operation has a large relative cost in actual computation time.

Images